The Global AI Buildout Needs Energy Transition Guardrails

The tensions around AI-related electricity demand are currently most visible in the United States and, increasingly, in Europe. The infrastructure models, utility arrangements, procurement practices, and policy precedents emerging in such major AI markets today are likely to influence how digital infrastructure develops globally.

Countries across Asia, Latin America, Africa, and the Gulf are increasingly positioning themselves as future digital infrastructure hubs, often while also navigating grid constraints, affordability pressures, water stress, energy access challenges, and climate goals. For many nations, interest in data centers and digital infrastructure is tied to legitimate development and economic priorities: expanding digital capacity, supporting domestic innovation, attracting investment, and reducing dependence on overseas cloud infrastructure.

In the best cases, large digital loads could help anchor additional clean energy and grid investment. AI tools themselves may also support the energy transition, including by improving renewable forecasting, optimizing grid operations, identifying faults faster, and making energy systems more efficient. Grid operators are already moving beyond theory to practice, testing AI-enabled tools to improve forecasting, accelerate interconnections, and manage increasingly complex power systems. But the opportunity is not simply “more data centers”—it is whether countries can capture the economic, digital, and infrastructure benefits of new investment without overreliance on fossil fuels, inequitable cost allocation, water stress, or environmental and public health harms.

These risks are especially acute in places where emerging digital infrastructure demand intersects with power reliability challenges, affordability pressures, or limited public oversight. The International Energy Agency (IEA) notes that in regions with frequent outages, maintaining data centers can require costly backup power systems. It also warns that in some Latin American and African countries, large-scale data center investments can sit alongside everyday energy scarcity, with remote communities still experiencing severe power shortages.

That is the heart of the global governance challenge. Without clear guardrails, large new data center loads can necessitate costly grid upgrades, additional generation, and backup power capacity. When these costs and impacts are passed on to the public, they can show up as higher electricity bills, increased public subsidies, or localized environmental harms. Countries should not be forced to choose between digital development and climate commitments, nor between attracting investment and protecting the affordability, public health, and rights of their citizens.

Governments have already made climate and environmental commitments through domestic laws, national energy plans, and international frameworks, including the Paris Agreement and emerging international efforts to align AI development with public interest goals. Digital infrastructure should be planned within those commitments, not treated as an exception to them. 

While data centers have long been consequential actors in corporate renewable energy markets, the infrastructure powering AI has, in a matter of years, become an increasingly important force shaping electricity demand, power-system planning, grid investment, land use, water use, and long-term energy choices.

The scale and speed of growth are significant. BloombergNEF identified 22.8 GW of IT or compute capacity currently under construction globally. While data center developers have served as foundational purchasers for renewable energy in many markets, they are now rapidly increasing spending and energy procurement, even as uncertainty remains about how much of the projected AI demand will materialize.

The impacts are especially sharp because the demand is geographically concentrated. A large data center can consume as much electricity as 100,000 households, and the largest ones under construction today could even consume far more. Data centers also tend to cluster near fiber networks, business centers, public incentives, favorable market rules, and available land or power. That means even if data centers remain a modest share of global electricity use, they can place heavy pressure on local grids and communities.

In some places, those pressures are already forcing hard choices. Utilities, system operators, and energy planners are revising load forecasts upwards. Regulators and public authorities are debating who should pay for new grid infrastructure. Governments are trying to attract AI investment while also meeting climate, affordability, reliability goals and minimizing environmental and public health impacts. And companies are competing to secure power at speed.

That is why data centers must now be understood as energy infrastructure, shaped by how new digital demand is planned, powered, and governed in the public interest.

As AI-related electricity demand grows, new fossil generation, especially gas, is increasingly being presented as the fastest or most reliable option. Reliability concerns are real. Data centers typically require high volumes of steady power. Grid connection queues are long, transmission upgrades can take years, and in some regions, the existing grid cannot accommodate major new loads without substantial investment. Where grid power is delayed or constrained, developers may also turn to on-site or backup generation, which can create additional air pollution and complicate fossil retirement plans if not carefully regulated.

But speed should not be confused with inevitability.

The risk is that AI demand, because it is perceived as urgent and economically attractive, gets treated as an exceptional load that can bypass normal public interest scrutiny. That can lead utilities, regulators, and policymakers toward a familiar shortcut: build more gas now and ask harder questions later.

The scale of this risk is already visible. BloombergNEF has identified 114 GW of gas-fired capacity across 115 projects under development or already online specifically to serve data center loads on-site. For perspective, Lawrence Berkeley National Laboratory’s latest review of interconnection queues found 136 GW of active natural gas capacity seeking grid connection in the United States. In other words, the announced global pipeline of on-site gas for data centers is approaching the scale of all active gas capacity currently seeking interconnection in the U.S. power sector. Nearly 90 percent of the announced on-site data center gas capacity is in the United States, with Texas alone accounting for a large share. 

That concentration matters globally. Other countries may observe these “speed to power” strategies and adopt similar approaches as their own digital infrastructure expands. However, much of this proposed gas buildout depends on highly optimistic data center growth projections, creating a clear risk of excess power capacity.

For countries trying to transition away from fossil fuels, new gas built around speculative or poorly scrutinized load growth can make the transition harder. Once approved and financed, gas plants and related infrastructure are designed to run for decades, and the contracts, cost-recovery expectations, and institutional pressures around them can make them difficult to retire even when cleaner alternatives become cheaper and more available.

Gas also does not guarantee speedy access to power. The IEA notes that turbine deliveries for new gas-fired power plants can be delayed by several years, which could push operation past 2030.

A rush to gas can also sidestep critical questions: Is the projected load real and properly timed? Are clean energy, battery storage, demand flexibility, and grid upgrades being accelerated to meet that demand? Are costs being shifted onto households or public budgets? And what local air pollution, climate, and public health impacts will communities be asked to absorb? Those questions should be answered before countries lock in new fossil infrastructure, not after.

Major technology companies have corporate climate commitments that have helped drive significant global investment in clean energy. In 2025, data center operators accounted for approximately half of all corporate clean power contracts globally, including power purchase agreements.

That progress matters. Clean power procurement has helped expand renewable energy markets and can send important signals to developers and investors. But procurement alone will not address the full suite of impacts that data centers can create for communities, energy systems, and the environment. A project can procure clean electricity and still strain local grids, increase costs for other electricity users, rely on polluting backup generation, put pressure on land and water resources, or move forward without meaningful public oversight.

When clean energy procurement is not incremental to existing supply, or cannot be delivered where and when the data center needs power, it can worsen local supply constraints and increase fossil fuel generation. A facility can also appear efficient within its own fence line while still increasing pressure on local grids, water resources, land, or air quality.

This is why the next phase of data center energy policy needs to move beyond corporate clean power purchases alone. Large new loads also need system planning: additional clean energy, available grid capacity, demand flexibility, storage, and siting decisions that reduce pressure on already constrained systems. The goal should be to add clean power while also modernizing grids and protecting other electricity users and communities from bearing the costs, pollution, and reliability impacts.

A more responsible path is possible, but it requires treating data center development as an energy planning and public interest decision, not simply a private procurement problem. The goal should be to establish baseline protections for every project while rewarding companies willing to meet higher standards. In practice, new data center load should be powered by additive clean energy, integrated into grid planning, and prevented from delaying fossil fuel retirement or shifting costs onto the public.

AI has real potential to support the clean energy transition, from improving renewable forecasting and grid operations to detecting faults faster and helping manage more complex electricity systems. The IEA finds that AI-based fault detection can reduce outage durations by 30 to 50 percent, and that AI-enabled tools could help unlock up to 175 GW of transmission capacity without building new lines.

But these benefits are not automatic. They depend on clear governance, enforceable standards, and public-private choices that align digital infrastructure with cleaner, more resilient, and more equitable energy systems. Without those guardrails, the public may be left paying for the consequences long after today’s AI models are obsolete.